On Gait-Based Identification of Persons During Winter Conditions

This paper investigates the effectiveness of gait-based identification across different seasons which is especially important for the northern territories. The study highlights the uniqueness of gait patterns, influenced by anatomical and physiological characteristics, and their potential for non-invasive identification methods. It addresses the critical question of whether a model trained on summer data can identify individuals in winter, revealing difficulties due to seasonal changes affecting gait patterns. The research employs a comprehensive dataset collected using smartphones equipped with accelerometers and gyroscopes, capturing various gait parameters across different terrains and seasons. The paper explores data filtering techniques and machine learning algorithms, including decision trees, k-nearest neighbors, and random forests, to analyze gait data. The results demonstrate the impact of seasonal variations on model performance, underscoring the need for diverse training datasets to enhance identification accuracy. The study’s findings suggest that while gait identification holds promise for various applications, its reliability is contingent on the inclusion of diverse seasonal data in model training. The study contributes to the understanding of the potential and limitations of gait analysis and contributes to the development of digital assistants for tracking human motor activity.